Autonomous Planetary Robots:

Fuzzy Control and Decision

Dr. Tarek A. Tutunji

2013

Reference

The following slides are captured from:

Tunstel, Seraji, and Howard “Soft Computing Approach to

Safe Navigation of Autonomous Planetary Rovers”

Control System using Soft Computing, Chapter 11

Introduction

National Aeronautics and Space Administration (NASA)

has been engaged in the conceptualization and

implementation of space flight missions to planet Mars

These planetary rovers must have mobility characteristics

that are sufficient for traversing rough and rugged terrain.

Moreover, due to the extreme remoteness of their

operating environment, Mars rovers must be capable of

operating autonomously and intelligently

Practical Issues

Autonomous rovers designed for planetary surface

exploration must be capable of point-to-point navigation

in the presence of varying obstacle distributions (rocks,

boulders, etc.), surface characteristics, and hazards.

The round trip communication time delay between Earth

and Mars, coupled with lack of frequent opportunities for

communication with landed resources on Mars, makes

direct control of a Mars rover all but impractical

Practical Issues

Constraints on power, computation, weight, and

communications bandwidth

Space flight projects require the use of proven, radiation-

hardened or otherwise space-flight-qualified electronics

that will survive and operate in the harsh temperature

and radiation extremes of space.

Need efficient algorithms

Overview

A fuzzy-logic-based reasoning and control framework is

described.

Also visual perception algorithms are used to realize a practical

rover navigation system.

Safe navigation system and soft computing techniques

have been applied to solve different aspects of the rover

navigation problem

Navigation System Overview

fuzzy inference systems are developed for navigation that

emulate human judgment and reasoning as derived from

off-road driving heuristics

Each component is implemented using fuzzy reasoning

with the exception of the low-level rover motion control

system,

Modular System Diagram

Fuzzy Behavior-Based Structure

The architectural design is based on the premise that

autonomous navigation functionality can be decomposed

into a finite number of special purpose task achieving and

decision making behaviors.

A behavior represents a mapping, from perceptions or

goals to actions or decisions, aimed at achieving a given

desired objective. That is, behaviors may be of two general

types: control behaviors and decision behaviors

Fuzzy Behavior-Based Structure

Fuzzy IF-THEN rules have the following form

IF x is Ci , THEN u is Ai

Input x refers to sensory data; u refers to motion control variables that influence rover translation and rotation.

The control variables serve as set points for low level classical PID motor controllers.

The control behaviors can be executed individually or concurrently to produce intelligent behavior for goal-directed navigation. Concurrent execution of fuzzy behaviors is facilitated by fuzzy decision-making modules

Fuzzy-Logic-Based Rover Health and Safety

1. Health and Safety Indicators

2. Stable Attitude Control

3. Traction Management

Health and Safety Indicators

Stable Attitude Control

Maintain upright stability

The rover is outfitted with a two-axis tilt sensor to measure pitch and roll

Simplest approach is to stop rover motion when either axis senses tilt beyond a critical threshold

Planetary robots are driven at low speeds (0.3 m/s)

Recommended safe speed for the rover is proportionately modulated in reaction to changes in attitude (pitch and roll)

Stable Attitude Control

Considering various off-road driving heuristics, a set of fuzzy rules are formulated

In addition to these rules, a crisp rule is applied to handle the extreme cases

Inputs: Pitch and Roll

Pitch is represented by five fuzzy sets:

{NEG-HIGH, NEG-LOW, ZERO, POS_LOW, POS-HIGH}

Roll is partitioned using three fuzzy sets:

{NEG, ZERO, POS}

Output: Velocity

Stable Attitude Control

Traction Management

Traction coefficient denoted by Ct

The rules

IF Ct is LOW, THEN v is SLOW.

IF Ct is MEDIUM, THEN v is MODERATE.

IF Ct is HIGH, THEN v is FAST.

Safe speeds recommended by the safety module are

compared to the strategic speed recommendations, and

the safest speed is issued as the commanded set point for

the motion control system.

Traction Management

Traction Management

Terrain-Based Fuzzy Navigation

An algorithm is applied to a pair of stereo camera images that determines the sizes and concentration of rocks/ditches in the viewable scene.

Rock sizes, Rs: {SMALL, LARGE},

Rock Concentration, Rc: {FEW, MANY}

Terrain roughness, β: {SMOOTH, ROUGH, ROCKY}

IF Rc is FEW AND Rs is SMALL, THEN β is SMOOTH.

IF Rc is FEW AND Rs is LARGE, THEN β is ROUGH.

IF Rc is MANY AND Rs is SMALL, THEN β is ROUGH.

IF Rc is MANY AND Rs is LARGE, THEN β is ROCKY.

Terrain-Based Fuzzy Navigation

Inputs : Terrain roughness b an terrain slope a

Output: Traversability

Strategic Fuzzy Navigation Behaviors

There are three motion behaviors:

Seek-goal

Traverse-terrain,

Avoid-obstacles.

In the final stage, the individual fuzzy recommendations

from the three behaviors are aggregated and defuzzified

to yield crisp control inputs

Seek-Goal Behavior

Navigate a rover on a natural terrain from a known initial

position to a user-specified goal position.

The rover control variables for this behavior are the

translational speed v and the rotational speed ω.

φ, called the heading error, is the relative angle by which

the rover needs to turn to face the goal directly

d, position error input (goal distance)

Seek-Goal Behavior

IF φ is GOAL-FAR LEFT, THEN ω is FAST-LEFT.

IF φ is GOAL-LEFT, THEN ω is SLOW-LEFT.

IF φ is GOAL-HEAD ON, THEN ω is ON-COURSE.

IF φ is GOAL-RIGHT, THEN ω is SLOW-RIGHT.

IF φ is GOAL-FAR RIGHT, THEN ω is FAST-RIGHT.

IF d is VERY NEAR OR φ is NOT GOAL-HEAD ON, THEN v is STOP.

IF d is NEAR AND φ is GOAL-HEAD ON, THEN v is SLOW.

IF d is FAR AND φ is GOAL-HEAD ON, THEN v is MODERATE.

IF d is VERY FAR AND φ is GOAL-HEAD ON, THEN v is FAST.

Traverse-Terrain Behavior

Fuzzy logic rules that use the fuzzy traversability index to

infer the vehicle turn rate and speed while moving on

natural terrain.

Visual sensor spans 180° that is partitioned into three 60°

sectors, namely: front, right, and left: τf, τr, τl,

The fuzzy rules for determining rover steering direction:

R: Right, L: Left, O: No turn

Traverse-Terrain Behavior

Avoid-Obstacle Behavior

Fuzzy-Behavior Fusion

Weighting factors s, t, and a represent the strengths by which the seek-goal, traverse-terrain, and avoid-obstacle recommendations are taken into account to compute the final control actions v and ω.

IF d is VERY NEAR, THEN s is HIGH.

IF d is NOT VERY NEAR, THEN s is NOMINAL.

IF d is NOT VERY NEAR AND df is NOT VERY NEAR, THEN t is HIGH.

IF d is VERY NEAR OR df is VERY NEAR, THEN t is NOMINAL.

IF d is NOT VERY NEAR, THEN a is HIGH.

IF d is VERY NEAR, THEN a is NOMINAL.

Fuzzy-Behavior Fusion

Conclusion

An autonomous planetary rover must be able to operate

intelligently with minimal interaction.

Robot navigation strategies based on fuzzy logic offer major

advantages over analytical methods.

First, the fuzzy rules that govern the robot motion are easily

understandable, intuitive, and emulate the human driver's experience.

Second, the tolerance of fuzzy logic to imprecision and uncertainties

in sensory data is particularly appealing for outdoor navigation

because of the inevitable inaccuracies in measuring and interpreting

the terrain quality data, such as slope and roughness.

Multiple fuzzy behaviors can be blended into a unified navigation

strategy